Anthropic signs multi-year CoreWeave deal to rent Nvidia cloud capacity
Anthropic signed a cloud rental deal with CoreWeave on Friday. The same day, Reuters reported it is exploring designing its own chips. The message in both facts is the same: there is not enough compute.
image from Gemini Imagen 4
Anthropic signed a multi-year agreement with CoreWeave to rent Nvidia-powered cloud capacity, joining a string of massive infrastructure deals including OpenAI's $11.9B and Meta's $21B contracts with the GPU cloud provider. Simultaneously, Reuters reported Anthropic is exploring designing its own AI chips—an effort estimated to cost ~$500M and still in early stages without a dedicated team—while also committing $50B to U.S. compute infrastructure through Google and Broadcom using custom TPUs. The three parallel strategies (renting, building custom silicon, and TPU partnerships) all address the same bottleneck: Anthropic's run-rate revenue has grown from $9B to $30B faster than its ability to build or buy the hardware to serve it.
- •CoreWeave is emerging as the dominant AI cloud rental platform, with $38B+ in announced deals in 2025 alone, signaling that cloud GPU renting remains the fastest path to compute access.
- •The simultaneous pursuit of custom chip design, TPU partnerships, and cloud rentals indicates AI labs are hedging across hardware strategies rather than committing to a single compute approach.
- •Anthropic's $30B run-rate revenue growing 3x in months while struggling to secure compute illustrates the supply-demand imbalance driving consolidation across the AI value chain.
